KR20170075127A - Method of simplified channel coding for processing packet loss recovery - Google Patents

Abstract

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to image processing, and more particularly, to a lightweight data recovery technique capable of maintaining media communication service quality such as video conferencing even when packet loss occurs on the Internet. According to an embodiment of the present invention, there is provided a method of generating a LT (Luby Transform) matrix composed of K x L elements using a first seed value, wherein at least some of the K x L elements Generating an HDPC (High Density Parity Check) matrix composed of H x K elements using a second seed value, generating an identity matrix composed of H x H elements, and generating the LT matrix, the HDPC matrix, And combining the unitary matrices to generate a combining matrix. The weighted channel coding method for packet loss recovery processing is provided.

Description

[0001] The present invention relates to a method and apparatus for packet loss recovery,

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to image processing, and more particularly, to a lightweight data recovery technique capable of maintaining media communication service quality such as video conferencing even when packet loss occurs on the Internet.

Currently, channel coding techniques are used for data transmission such as file transfer or video streaming. In the case of Real-time Transport Protocol (RTP) packets, a technology that can recover lost packets by using ALFEC (Application Layer Forward Error Correction) is used because it does not use retransmission. In particular, fountain code techniques such as Raptor code and RaptorQ code can be used to create as many overhead packets as necessary.

FIG. 1 is a diagram illustrating a RaptorQ encoding and decoding procedure, and FIG. 2 is a diagram illustrating the number of LDPC and HDPC symbols in a RaptorQ code.

Referring to FIG. 1, RaptorQ code technology improves the recovery rate of lost packets by using LDPC (Low Density Parity Check), HDPC (High Density Parity Check) and LT code (Luby Transform Code). Here, LDPC and HDPC correspond to pre-code and LT code is used as fountain code to generate additional symbols. A combination of LDPC, HDPC, and LT codes is required to construct a matrix of RaptorQ codes and to reverse the matrix during encoding / decoding. The inverse procedure of Matrix is to use RaptorQ code as systematic code. Systematic code means that the input symbol remains in the output symbol in its original form. The larger the matrix size of the RaptorQ code is, the larger the amount of computation required in the matrix is, and the higher the LT code degree is, the more the computational complexity is increased. Here, "Degree" means the number of 1s in the vector constituting the LT matrix. Therefore, it can be said that the smaller the degree of the LT code is, the smaller the calculation amount of RaptorQ code is while the whole matrix is small.

The current RaptorQ code increases the size of LDPC and HDPC matrices as the number of symbols increases. Referring to FIG. 2, although the size of the HDPC is very small, the size of the LDPC is larger than that of the HDPC. Also, the ratio of the LDPC matrix to the entire matrix is higher when the number of symbols is small.

The present invention reduces coding and decoding processing delay to enable application using high-quality real-time video conferencing or a small number of symbols over HD class (720p) or higher in a system capable of limited calculation such as a smart phone, and is similar to RaptorQ code The goal is to provide packet loss recovery performance.

According to an embodiment of the present invention, there is provided a method of generating a LT (Luby Transform) matrix composed of K x L elements using a first seed value, wherein at least some of the K x L elements Generating an HDPC (High Density Parity Check) matrix composed of H x K elements using a second seed value, generating an identity matrix composed of H x H elements, and generating the LT matrix, the HDPC matrix, And combining the unitary matrices to generate a combining matrix. The weighted channel coding method for packet loss recovery processing is provided.

According to another embodiment of the present invention, there is provided a method for transmitting an ESI symbol, comprising the steps of: distinguishing an original symbol and an additional symbol with an Encoding Symbol ID (ESI) value of a received symbol; storing a lost ESI number among original symbols; Generating an association matrix A including an LT matrix using the ESI of the received symbol if the association matrix A is inverse; generating an intermediate symbol if the association matrix A is reversible; Generating a 1 x L vector using P = (log Lp + d) / Lp, and recovering the lost symbol using the generated vector and the intermediate symbol. A coding method is provided.

In one embodiment, the size of the entire matrix is reduced by removing the LDPC and using the new LT matrix, but the RaptorQ code level recovery performance is maintained. Therefore, when the number of symbols is small, for example, below about 500, the average degree of the LT code can be smaller than RaptorQ code.

In one embodiment, the LDPC code is removed from the matrix configuration used in the RaptorQ code, and the degree distribution of the LT code is changed. As a result, the size of the entire matrix is reduced and the amount of calculation is reduced when the intermediate symbol is generated. Here, the average degree of the proposed LT code shows an average value of log (K + H) + 2. The average degree tends to be smaller than the average degree of RaptorQ code when the number K of original symbols is less than about 500, for example. If the average degree is small, it is more efficient in terms of calculation amount in recovery of additional symbol generation / loss symbols.

According to the present invention, high-quality real-time video conferencing of higher than HD grade or applications using a small number of symbols can be performed in a system capable of limited calculation such as a smart phone.

Hereinafter, the present invention will be described with reference to the embodiments shown in the accompanying drawings. For the sake of clarity, throughout the accompanying drawings, like elements have been assigned the same reference numerals. It is to be understood that the present invention is not limited to the embodiments illustrated in the accompanying drawings, but may be embodied in many other specific forms without departing from the spirit or essential characteristics thereof.
Figure 1 is an exemplary diagram illustrating a RaptorQ encoding and decoding procedure.
2 is a diagram showing the number of symbols of LDPC and HDPC of the RaptorQ code.
FIG. 3 is an exemplary diagram illustrating a configuration of a combined matrix A according to an embodiment of the present invention.
4 is a flowchart illustrating an exemplary method of generating the combining matrix A shown in FIG.
5 is a flowchart illustrating an exemplary procedure for generating an additional symbol in an encoder.
FIG. 6 is a flowchart illustrating a process of recovering a lost symbol on a receiving side.

While the present invention has been described in connection with certain exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover various modifications and similarities. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

FIG. 3 is an exemplary diagram illustrating a configuration of a combined matrix A according to an embodiment of the present invention.

Referring to FIG. 3, the combining matrix A is composed of an HDPC matrix 101 of H x K size, a unit matrix I _H 102 of H x H size, and an LT matrix 103 of K x L. The HDPC matrix 101 is a random matrix composed of numbers between 1 and 255, and operates on Galois Field 256 (GF256) in matrix operation. The LT matrix 103 has a structure in which the number of 1's is less distributed, and the distribution of 1 can be determined, for example, according to the probability equation P = (log Lp + d) / Lp. Thus, the average degree of the LT matrix 103 may be logL + d. Where L is the sum of the number K of original symbols and the number H of HDPC symbols (L = K + H), and d is a constant equal to or greater than zero. The probability that the inverse of the combining matrix A is possible can be expressed as a product of the probability that the rank of the K x L LT matrix GF2 is K and the rank of the H x H unit matrix I _H (GF256) is H. In one embodiment of the present invention, when the number of symbols is 1000 or less, the d value is set to 2, and when the H value is set to 10, a loss recovery probability similar to RaptorQ code can be obtained.

4 is a flowchart illustrating an exemplary method of generating the combining matrix A shown in FIG.

Referring to FIG. 4, the process of generating the LT matrix is as follows. In step 401, a seed value is set, and in step 402, a 1 x L vector may be generated using, for example, the probability P = (log Lp + d) / Lp. In this case, the probability P means the probability that each element of the 1 x L vector becomes 1. In step 403, it is determined whether the number of rows in the LT matrix is smaller than K. If the number of rows in the LT matrix is smaller than K, the steps 401 and 402 are repeated to generate the LT matrix.

The generation process of the HDPC matrix is as follows. In step 405, a new seed value is set, and in step 406, a 1 x K vector is generated. Here, each element of the 1 x K vector is generated to have a value ranging from 1 to 255 randomly. In step 406, the generated 1 x K vector is added to the HDPC matrix x. In step 407, it is determined whether the number of rows in the HDPC matrix is smaller than H. If the number of rows in the HDPC matrix is smaller than H, the steps 405 and 406 are repeated until the number of rows in the HDPC matrix becomes H.

Also, at step 408, an identity matrix I _H of size H x _H is generated.

In step 404, the generated LT matrix, HDPC matrix, and unit matrix I _H are combined in the form as shown in FIG. 3 to generate a combining matrix A. The join matrix A becomes an L x L matrix and can be used when the rank of the join matrix A is L in the transmitter. If the rank of the join matrix A is not L, then the join matrix A must be constructed using different seed values.

5 is a flowchart illustrating an exemplary procedure for generating an additional symbol in an encoder.

Referring to FIG. 5, in step 501, a seed value is incremented by one more than K_i used in step 401 when generating an additional symbol. In step 502, a K + i-th vector is generated with the same probability as in step 402. In step 503, the vector generated in step 502 and the intermediate symbol in step 506 are input to generate additional symbols. At step 505, it is determined if more symbols are needed, and if necessary, returns to step 501 and ends when it is no longer needed. The method of generating additional symbols is the same as RaptorQ code and can be calculated as C x R_v. Here, R_v is a redundant symbol vector, and C is an intermediate symbol.

FIG. 6 is a flowchart illustrating a process of recovering a lost symbol on a receiving side.

Referring to FIG. 6, in step 601, an ESI (Encoding Symbol ID) value of a received symbol is distinguished from an original symbol or an additional symbol, and the lost ESI number is stored in the original symbol. Here, ESI is information for uniquely identifying an encoding system associated with a source block for transmission / reception, and ESI is the same as that used in RFC6330.

In step 602, it is determined whether the number of received total symbols R _count is greater than or equal to K. [ If the total number of symbols R _count is greater than K, the process proceeds to step 603; otherwise, the process returns to step 401.

In step 603, it is determined whether or not the original symbol is lost among the received symbols. If there is no loss, there is no symbol to be recovered. If there is a lost symbol, the LT matrix is generated using the ESI of the received symbol and a combining matrix A of the type shown in FIG. 3 is generated. At this time, the HDPC matrix and the unit matrix are the same as the matrix generated in FIG.

In step 605, it is determined whether or not the generated combining matrix A can be inverted. If the combining matrix A is not reversible, go back to step 601 to receive additional symbols.

In step 607, an Intermediate symbol is generated, and in step 606, the lost ESI number stored in step 601 is multiplied by a 1 x L vector using the probability P = (log Lp + d) / Lp, , And recover the lost symbol using the generated vector and the intermediate symbol.

It will be understood by those skilled in the art that the foregoing description of the present invention is for illustrative purposes only and that those of ordinary skill in the art can readily understand that various changes and modifications may be made without departing from the spirit or essential characteristics of the present invention. will be. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive.

It is intended that the present invention covers the modifications and variations of this invention provided they come within the scope of the appended claims and their equivalents. .

Claims (1)

Generating a LT (Luby Transform) matrix consisting of K x L elements using a first seed value, wherein at least some of the K x L elements are 1 according to a predetermined probability;
Generating an HDPC (High Density Parity Check) matrix composed of H x K elements using the second seed value;
Generating a unit matrix composed of H x H elements; And
And combining the LT matrix, the HDPC matrix, and the unitary matrix to generate a combining matrix.

Images